Linking Gauge R&R to FMEA Detection Ratings: When Your Measurement System Caps the Score

A final-inspection gauge check sits in your PFMEA rated Detection = 2 — “controls almost certain to detect.” The team assigned it because the operator measures every part with a digital caliper against a tight bore tolerance. Then the measurement systems analysis comes back: the gauge study shows 31% Gauge R&R. The caliper cannot reliably tell a conforming bore from a nonconforming one, which means the Detection = 2 you wrote down is fiction. Your detection rating can never be better than the measurement system that produces it allows.

This is the link practitioners skip: the Detection rating assumes the control works, and Gauge R&R is the evidence of whether it actually does. This walkthrough computes a real Gauge R&R study, reads its two key outputs, and shows where the result caps the Detection score you are allowed to claim.

The two numbers a Gauge R&R study gives you

A crossed Gauge R&R study (multiple operators, multiple parts, repeated trials) decomposes the variation an operator sees into measurement-system variation and part-to-part variation. Two outputs matter for FMEA Detection.

The first is %GRR — how much of the observed variation comes from the gauge and the operators rather than from real differences between parts:

$$\%GRR = \frac{GRR}{TV} \times 100, \quad GRR = \sqrt{EV^2 + AV^2}, \quad TV = \sqrt{GRR^2 + PV^2}$$

where EV is equipment variation (repeatability — the same operator, same part), AV is appraiser variation (reproducibility — different operators), and PV is part-to-part variation. TV is the total study variation. Per the AIAG Measurement Systems Analysis reference manual, %GRR under 10% is acceptable, 10–30% is conditionally acceptable depending on the application and tolerance, and above 30% is unacceptable. ASQ’s Gage R&R guidance uses the same acceptance bands.

Key Formula $$ndc = 1.41 \times \frac{PV}{GRR}$$

The second output is the number of distinct categories (ndc) — how many non-overlapping groups the measurement system can actually resolve across the part range. AIAG MSA guidance treats ndc of 5 or more as the minimum for a variable-data gauge; below 5, the system behaves more like an attribute (pass/fail) gauge than a measuring one.

A worked study, end to end

Worked Example A bore-diameter caliper study returns: repeatability EV = 0.04 mm, reproducibility AV = 0.03 mm, part-to-part PV = 0.15 mm. Work the two outputs:

First the combined Gauge R&R, then the total study variation:

$$GRR = \sqrt{0.04^2 + 0.03^2} = \sqrt{0.0025} = 0.05 \text{ mm}$$ $$TV = \sqrt{0.05^2 + 0.15^2} = \sqrt{0.025} = 0.158 \text{ mm}$$

Now the two decision numbers:

$$\%GRR = \frac{0.05}{0.158} \times 100 = 31.6\%$$ $$ndc = 1.41 \times \frac{0.15}{0.05} = 4.23 \rightarrow 4$$

Both fail. The 31.6% GRR sits above the 30% unacceptable line, and ndc of 4 falls below the threshold of 5. The caliper, as used by these operators on these parts, cannot separate good bores from marginal ones. The detection control built on it does not detect — it guesses.

Sanity check the result

Before you act on a study, confirm it is internally consistent. PV (0.15) dominates GRR (0.05) in absolute terms, yet %GRR is still high — that is the warning. It tells you the part range in the study was narrow relative to the gauge noise, which is exactly the condition that collapses ndc. If your study parts span the real production range and %GRR is still above 30%, the gauge is the problem, not the sample. Re-running the study with wider parts only papers over a gauge that cannot resolve the tolerance.

How the result caps your Detection rating

There is no AIAG-VDA table that converts %GRR into a Detection score — Detection is rated against the control’s ability to detect the failure mode, and the handbook leaves the measurement-capability judgment to the team. But the cap follows directly from the logic: a control cannot detect more reliably than its measurement system can resolve. The practitioner rule below is engineering judgment, not a published standard, and it keeps optimistic Detection ratings honest.

Common Mistake Writing Detection = 1–3 for a gauge-based check without ever looking at its Gauge R&R. A Detection = 1 on a control with 31% GRR is the most dangerous line in the FMEA: it looks like a covered risk and behaves like an open one.

A defensible cap on the Detection rating for a variable-data gauging control:

%GRR < 10% and ndc ≥ 5 — the measurement system supports whatever low Detection rating the control logic earns (automated 100% gauging with error-proofing can reach Detection 1–3).
%GRR 10–30% and ndc ≥ 5 — adequate but not capable for tight tolerances; do not rate Detection better than roughly 4–5 for the characteristic this gauge measures.
%GRR > 30% or ndc < 5 — the system is not a capable variable gauge; the failure mode should carry a high Detection rating (roughly 7–10) no matter how thorough the check feels, until the measurement system is fixed.

The worked study above lands in the third band, so the bore-diameter failure mode keeps a Detection near 7–8 — and the real recommended action is to fix the measurement system, not to add another inspection on top of one that cannot see. This is the same data-to-rating discipline used on the occurrence axis; the same logic that lets you convert Cpk and process-capability data into an occurrence rating applies in reverse for detection, and it sharpens how you evaluate current controls and assign detection scores across the whole worksheet.

Where this breaks

The %GRR rule is for variable-data measurement. Two common cases need a different method. Attribute (go/no-go) gauges and visual checks have no continuous reading, so their capability is assessed with an attribute agreement analysis — Cohen’s kappa or the AIAG signal-detection approach — not %GRR; a poor kappa caps Detection the same way a poor %GRR does. Destructive tests (weld pull, burst) cannot remeasure the same part, so they need a nested or expanded study design rather than the standard crossed study. And teams sometimes argue %GRR against the tolerance (%P/T) instead of against total variation; for a Detection judgment, the study-variation view above is the one that tells you whether the gauge can distinguish parts at all.

Turn the capped rating into the right action

Once the Detection cap is honest, the S, O, and D values feed the risk math. If you are running AIAG-VDA Action Priority rather than a raw RPN, the RPN and Action Priority calculator takes the capped Detection alongside Severity and Occurrence and returns the AP level — so a measurement system that caps Detection at 8 shows up as elevated priority instead of hiding behind a fictional 2.

The next time someone proposes a low Detection rating for a gauge-based control, ask for the Gauge R&R study before you write the number. If the measurement system cannot resolve the characteristic, the detection is not real — and the FMEA is hiding a risk it claims to have closed.